Re: st: Graph bar with label wtih % symbol at the top of each bar

[David Hoaglin <dchoaglin@gmail.com>]

Nick,

You make a number of interesting points.  I don't have time now to
respond to all of them.

The main idea is to let the bar graph serve its function, showing
relative size, with as little clutter as possible.  If only three
numbers are involved, and they are needed, they can be given in text;
no need for a table.

I had in mind an analysis along the lines of what Ed Tufte did in
discussing the data-ink ratio in The Visual Display of Quantitative
Information (Graphics Press, 1983).  To illustrate that "redundant
data-ink depicts the same number over and over," he used a single
shaded bar with the number 35.9 above it: "The labeled, shaded bar ...
unambiguously locates the altitude in six separate was (any five of
the six can be erased and the sixth will still indicate the height):
as the (1) height of the left line, (2) height of shading, (3) height
of right line, (4) position of top horizontal line, (5) position (not
content) of number at bar's top, and (6) the number itself.  That is
more ways than are needed."

By contemporary standards, Brinton's emphasis on including the
numerical data from which the chart was made would be considered bad
advice.  Graphs with divided bars are usually to be avoided.

John Tukey did indeed mix graphical and numerical elements.  I could
probably find a few examples of graphs that use numbers as plotting
symbols (and others that use short names, such as the months of the
year).  I don't recall any examples of bars with numbers atop them.
As a "semi-graphical display," the stem-and-leaf display is in a
different category.

Ed Tufte acknowledged the inspiration and encouragement of Fred
Mosteller and John Tukey.

David Hoaglin

On Fri, May 11, 2012 at 6:49 AM, Nick Cox <n.j.cox@durham.ac.uk> wrote:
> On this point you can find authorities to support any prejudice.
>
> If I dissect this attitude, I find these thoughts -- in others; I will let David spell out his own arguments if they differ --
>
> 1. Giving numbers too is redundant, as it's the job of bar height or length to denote magnitude.
>
> 2. Giving numbers too makes a graph too busy, i.e. too detailed or complicated, to be really effective.
>
> 3. Figures are figures, and tables are tables, and ne'er the twain shall meet.
>
> On #1, I see the point, but numbers too give the information in a different form and that can be helpful or even useful as well. "60% of Stata users do that! I didn't know the figure was so high" or whatever. The number can be read off the graph at least approximately, to be sure, but sometimes the exact value is the take-home message that means most to the audience, and giving it to them is a nice thing to do. In teaching I use a bar chart with continent areas; even geographers need to be told, or reminded, quite how big Antarctica is, for example, and giving them a number too is reasonable.
>
> On #2, I think the answer is try it and see. A bar chart with three bars, the example that started this thread, could seem bare or even trivial without some annotation, but if the numbers are important, they are likely to enhance the graph. A bar chart with thirty bars could well look too crowded with the numbers too, and it is less likely that they will be legible.
>
> On #1 and #2, I do think there is a good case if you show numeric values, you don't need a numeric axis with numeric labels too.
>
> On #3, I think that is a convention to be ignored or flouted whenever it gets in the way of understanding the data. David and I share John Tukey as a statistical hero, and Tukey urged the mixing of graph and table ideas whenever it helped. Stem-and-leaf plots are perhaps the most obvious example, although not necessarily the best.
>
> I was looking again recently at
>
> Willard C[ope] Brinton. 1914. Graphic methods for presenting facts. New York: Engineering Magazine Company.
>
> Brinton was the Edward Tufte of his day, even to the extent of self-publishing his magnum opus
>
> Brinton, W.C. 1939.  Graphic presentation.  New York: Brinton Associates.  http://www.archive.org/stream/graphicpresentat00brinrich
>
> Much of what he said in 1914 bears repetition. His advice on this point seems quite well balanced:
>
> "Whenever possible, include in the chart the numerical data from which the chart was made." (p.361)
>
> "If numerical data cannot be included in the chart, it is well to show the numerical data in tabular form accompanying the chart." (p.362)
>
> P.S. a few more highlights from Brinton (1914):
>
> "The circle with sectors is not a desirable form of presentation" (p.6)
> -- as not so flexible as divided horizontal bar
>
> [the term "pie chart" was introduced later]
>
> "the horizontal arrangement lends itself more readily to the use of type so that the reader may read type statements without having to turn the book" (p.7) (cf. p.41)
>
> "Graphic comparisons, wherever possible, should be made in one dimension only." (p.22) (e.g. the use of bars of different lengths)
>
> "Avoid using areas or volumes when representing quantities.
> Presentations read from one dimension only are the least likely to be misinterpreted." (p.361)
>
> "The general arrangement of a chart should proceed from left to right."
> (p.361)
>
> Nick
> n.j.cox@durham.ac.uk

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/